The invention provides a depth camera-based invasion detection method. The method performs detection by the following steps of firstly setting a monitoring range in a camera range by a user; initializing a background pixel number in the preset monitoring range, and recording the background pixel number as num0; recording an invasion state as state, and initializing state to be 0, wherein when state is equal to 0, it is represented that no persons exist, and when state is equal to 1, it is represented that persons exist; for each frame of distance image data, performing statistics on a pixel number in a preset space range, and recording the pixel number as numi; if numi is less than a sum of num0 and thr1, executing the following operation: Num0=(A*num0+B*numi)/(A+B), wherein thr1 is a first threshold of judgment and reflects a noise fluctuation condition of the background pixel number, and A and B are used for adjusting the updating speed of the background pixel number; and if numi is greater than a sum of num0 and thr2, executing the following operation: setting State to be 1, wherein thr2 is a second threshold of judgment, and misjudgment of small animal invasion can be eliminated by adjusting thr2; or otherwise, keeping State unchanged. When invasion exists in the preset monitoring range, a point number is changed, and whether invasion of a person exists or not is judged according to a change situation. Compared with a mainstream invasion detection method at present, the depth camera-based invasion detection method has the advantages of basically no influence of ambient light and high identification accuracy.